def main():
args = parse_args()
config = InferenceConfig.from_config_dir(args.exp_dir)
config.test_fn = args.test_file
dataset = InferenceDataset(config)
model = Seq2seqInferenceModel(config, dataset)
model.run_inference(outfile=args.output_file)
def __init__(self,config=None):
if config == None:
config = InferenceConfig()
dir_path = os.path.dirname(os.path.realpath(__file__))
file = open(os.path.join(dir_path, config.MODEL_DIR, config.MODEL_FILE), 'rb')
self._svm_model = pickle.load(file)
file.close()
def run(filepath, is_gt, visualize):
# FIlepath: Path to write file csv for test, else path to gt csv
print(visualize)
# select trained model
dir_names = next(os.walk(MODEL_DIR))[1]
print(MODEL_DIR)
key = config.NAME.lower()
dir_names = filter(lambda f: f.startswith(key), dir_names)
dir_names = sorted(dir_names)
print(dir_names)
if not dir_names:
import errno
raise FileNotFoundError(
errno.ENOENT,
"Could not find model directory under {}".format(MODEL_DIR))
fps = []
# Pick last directory
for d in dir_names:
dir_name = os.path.join(MODEL_DIR, d)
# Find the last checkpoint
checkpoints = next(os.walk(dir_name))[2]
checkpoints = filter(lambda f: f.startswith("mask_rcnn"), checkpoints)
checkpoints = sorted(checkpoints)
if not checkpoints:
print('No weight files in {}'.format(dir_name))
else:
checkpoint = os.path.join(dir_name, checkpoints[-1])
fps.append(checkpoint)
model_path = sorted(fps)[-1]
print('Found model {}'.format(model_path))
inference_config = InferenceConfig()
# Recreate the model in inference mode
model = modellib.MaskRCNN(mode='inference',
config=inference_config,
model_dir=MODEL_DIR)
# Load trained weights (fill in path to trained weights here)
assert model_path != "", "Provide path to trained weights"
print("Loading weights from ", model_path)
model.load_weights(model_path, by_name=True)
# Get filenames of test dataset DICOM images
test_image_fps = get_dicom_fps(test_dicom_dir)
# Write predictions to file
predict(model, test_image_fps, filepath=filepath, is_gt=is_gt, visualize=visualize)
def evaluate(options):
config = InferenceConfig(options)
config.FITTING_TYPE = options.numAnchorPlanes
if options.dataset == '':
dataset = PlaneDataset(options,
config,
split='test',
random=False,
load_semantics=False)
elif options.dataset == 'occlusion':
config_dataset = copy.deepcopy(config)
config_dataset.OCCLUSION = False
dataset = PlaneDataset(options,
config_dataset,
split='test',
random=False,
load_semantics=True)
elif 'nyu' in options.dataset:
dataset = NYUDataset(options, config, split='val', random=False)
elif options.dataset == 'synthia':
dataset = SynthiaDataset(options, config, split='val', random=False)
elif options.dataset == 'kitti':
camera = np.zeros(6)
camera[0] = 9.842439e+02
camera[1] = 9.808141e+02
camera[2] = 6.900000e+02
camera[3] = 2.331966e+02
camera[4] = 1242
camera[5] = 375
dataset = InferenceDataset(
options,
config,
image_list=glob.glob('../../Data/KITTI/scene_3/*.png'),
camera=camera)
elif options.dataset == '7scene':
camera = np.zeros(6)
camera[0] = 519
camera[1] = 519
camera[2] = 320
camera[3] = 240
camera[4] = 640
camera[5] = 480
dataset = InferenceDataset(
options,
config,
image_list=glob.glob('../../Data/SevenScene/scene_3/*.png'),
camera=camera)
elif options.dataset == 'tanktemple':
camera = np.zeros(6)
camera[0] = 0.7
camera[1] = 0.7
camera[2] = 0.5
camera[3] = 0.5
camera[4] = 1
camera[5] = 1
dataset = InferenceDataset(
options,
config,
image_list=glob.glob('../../Data/TankAndTemple/scene_4/*.jpg'),
camera=camera)
elif options.dataset == 'make3d':
camera = np.zeros(6)
camera[0] = 0.7
camera[1] = 0.7
camera[2] = 0.5
camera[3] = 0.5
camera[4] = 1
camera[5] = 1
dataset = InferenceDataset(
options,
config,
image_list=glob.glob('../../Data/Make3D/*.jpg'),
camera=camera)
elif options.dataset == 'popup':
camera = np.zeros(6)
camera[0] = 0.7
camera[1] = 0.7
camera[2] = 0.5
camera[3] = 0.5
camera[4] = 1
camera[5] = 1
dataset = InferenceDataset(
options,
config,
image_list=glob.glob('../../Data/PhotoPopup/*.jpg'),
camera=camera)
elif options.dataset == 'cross' or options.dataset == 'cross_2':
image_list = [
'test/cross_dataset/' + str(c) + '_image.png' for c in range(12)
]
cameras = []
camera = np.zeros(6)
camera[0] = 587
camera[1] = 587
camera[2] = 320
camera[3] = 240
camera[4] = 640
camera[5] = 480
for c in range(4):
cameras.append(camera)
continue
camera_kitti = np.zeros(6)
camera_kitti[0] = 9.842439e+02
camera_kitti[1] = 9.808141e+02
camera_kitti[2] = 6.900000e+02
camera_kitti[3] = 2.331966e+02
camera_kitti[4] = 1242.0
camera_kitti[5] = 375.0
for c in range(2):
cameras.append(camera_kitti)
continue
camera_synthia = np.zeros(6)
camera_synthia[0] = 133.185088
camera_synthia[1] = 134.587036
camera_synthia[2] = 160.000000
camera_synthia[3] = 96.000000
camera_synthia[4] = 320
camera_synthia[5] = 192
for c in range(2):
cameras.append(camera_synthia)
continue
camera_tanktemple = np.zeros(6)
camera_tanktemple[0] = 0.7
camera_tanktemple[1] = 0.7
camera_tanktemple[2] = 0.5
camera_tanktemple[3] = 0.5
camera_tanktemple[4] = 1
camera_tanktemple[5] = 1
for c in range(2):
cameras.append(camera_tanktemple)
continue
for c in range(2):
cameras.append(camera)
continue
dataset = InferenceDataset(options,
config,
image_list=image_list,
camera=cameras)
elif options.dataset == 'selected':
image_list = glob.glob('test/selected_images/*_image_0.png')
image_list = [
filename for filename in image_list
if '63_image' not in filename and '77_image' not in filename
] + [
filename for filename in image_list
if '63_image' in filename or '77_image' in filename
]
camera = np.zeros(6)
camera[0] = 587
camera[1] = 587
camera[2] = 320
camera[3] = 240
camera[4] = 640
camera[5] = 480
dataset = InferenceDataset(options,
config,
image_list=image_list,
camera=camera)
elif options.dataset == 'comparison':
image_list = [
'test/comparison/' + str(index) + '_image_0.png'
for index in [65, 11, 24]
]
camera = np.zeros(6)
camera[0] = 587
camera[1] = 587
camera[2] = 320
camera[3] = 240
camera[4] = 640
camera[5] = 480
dataset = InferenceDataset(options,
config,
image_list=image_list,
camera=camera)
elif 'inference' in options.dataset:
image_list = glob.glob(options.customDataFolder +
'/*.png') + glob.glob(options.customDataFolder +
'/*.jpg')
if os.path.exists(options.customDataFolder + '/camera.txt'):
camera = np.zeros(6)
with open(options.customDataFolder + '/camera.txt', 'r') as f:
for line in f:
values = [
float(token.strip()) for token in line.split(' ')
if token.strip() != ''
]
for c in range(6):
camera[c] = values[c]
continue
break
pass
else:
camera = [
filename.replace('.png', '.txt').replace('.jpg', '.txt')
for filename in image_list
]
pass
dataset = InferenceDataset(options,
config,
image_list=image_list,
camera=camera)
pass
print('the number of images', len(dataset))
dataloader = DataLoader(dataset,
batch_size=1,
shuffle=False,
num_workers=1)
epoch_losses = []
data_iterator = tqdm(dataloader, total=len(dataset))
specified_suffix = options.suffix
with torch.no_grad():
detectors = []
for method in options.methods:
if method == 'w':
options.suffix = 'pair_' + specified_suffix if specified_suffix != '' else 'pair'
detectors.append(('warping',
PlaneRCNNDetector(options,
config,
modelType='pair')))
elif method == 'b':
options.suffix = specified_suffix if specified_suffix != '' else ''
detectors.append(('basic',
PlaneRCNNDetector(options,
config,
modelType='pair')))
elif method == 'o':
options.suffix = 'occlusion_' + specified_suffix if specified_suffix != '' else 'occlusion'
detectors.append(('occlusion',
PlaneRCNNDetector(options,
config,
modelType='occlusion')))
elif method == 'p':
detectors.append(
('planenet', PlaneNetDetector(options, config)))
elif method == 'e':
detectors.append(
('planerecover', PlaneRecoverDetector(options, config)))
elif method == 't':
if 'gt' in options.suffix:
detectors.append(
('manhattan_gt',
TraditionalDetector(options, config, 'manhattan_gt')))
else:
detectors.append(
('manhattan_pred',
TraditionalDetector(options, config,
'manhattan_pred')))
pass
elif method == 'n':
options.suffix = specified_suffix if specified_suffix != '' else ''
detectors.append(('non_planar',
DepthDetector(options,
config,
modelType='np')))
elif method == 'r':
options.suffix = specified_suffix if specified_suffix != '' else ''
detectors.append(('refine',
PlaneRCNNDetector(options,
config,
modelType='refine')))
elif method == 's':
options.suffix = specified_suffix if specified_suffix != '' else ''
detectors.append(
('refine_single',
PlaneRCNNDetector(options,
config,
modelType='refine_single')))
elif method == 'f':
options.suffix = specified_suffix if specified_suffix != '' else ''
detectors.append(('final',
PlaneRCNNDetector(options,
config,
modelType='final')))
pass
continue
pass
if not options.debug:
for method_name in [detector[0] for detector in detectors]:
os.system('rm ' + options.test_dir + '/*_' + method_name + '.png')
continue
pass
all_statistics = []
for name, detector in detectors:
statistics = [[], [], [], []]
for sampleIndex, sample in enumerate(data_iterator):
if options.testingIndex >= 0 and sampleIndex != options.testingIndex:
if sampleIndex > options.testingIndex:
break
continue
input_pair = []
camera = sample[30][0].cuda()
for indexOffset in [
0,
]:
images, image_metas, rpn_match, rpn_bbox, gt_class_ids, gt_boxes, gt_masks, gt_parameters, gt_depth, extrinsics, planes, gt_segmentation = sample[
indexOffset +
0].cuda(), sample[indexOffset + 1].numpy(), sample[
indexOffset +
2].cuda(), sample[indexOffset + 3].cuda(), sample[
indexOffset +
4].cuda(), sample[indexOffset + 5].cuda(), sample[
indexOffset +
6].cuda(), sample[indexOffset + 7].cuda(
), sample[indexOffset + 8].cuda(), sample[
indexOffset + 9].cuda(), sample[
indexOffset +
10].cuda(), sample[indexOffset +
11].cuda()
masks = (
gt_segmentation == torch.arange(gt_segmentation.max() +
1).cuda().view(-1, 1,
1)).float()
input_pair.append({
'image': images,
'depth': gt_depth,
'bbox': gt_boxes,
'extrinsics': extrinsics,
'segmentation': gt_segmentation,
'camera': camera,
'plane': planes[0],
'masks': masks,
'mask': gt_masks
})
continue
if sampleIndex >= options.numTestingImages:
break
with torch.no_grad():
detection_pair = detector.detect(sample)
pass
if options.dataset == 'rob':
depth = detection_pair[0]['depth'].squeeze().detach().cpu(
).numpy()
os.system('rm ' +
image_list[sampleIndex].replace('color', 'depth'))
depth_rounded = np.round(depth * 256)
depth_rounded[np.logical_or(depth_rounded < 0,
depth_rounded >= 256 * 256)] = 0
cv2.imwrite(
image_list[sampleIndex].replace('color', 'depth').replace(
'jpg', 'png'), depth_rounded.astype(np.uint16))
continue
if 'inference' not in options.dataset:
for c in range(len(input_pair)):
evaluateBatchDetection(
options,
config,
input_pair[c],
detection_pair[c],
statistics=statistics,
printInfo=options.debug,
evaluate_plane=options.dataset == '')
continue
else:
for c in range(len(detection_pair)):
np.save(
options.test_dir + '/' + str(sampleIndex % 500) +
'_plane_parameters_' + str(c) + '.npy',
detection_pair[c]['detection'][:, 6:9])
np.save(
options.test_dir + '/' + str(sampleIndex % 500) +
'_plane_masks_' + str(c) + '.npy',
detection_pair[c]['masks'][:, 80:560])
continue
pass
if sampleIndex < 30 or options.debug or options.dataset != '':
visualizeBatchPair(options,
config,
input_pair,
detection_pair,
indexOffset=sampleIndex % 500,
suffix='_' + name + options.modelType,
write_ply=options.testingIndex >= 0,
write_new_view=options.testingIndex >= 0
and 'occlusion' in options.suffix)
pass
if sampleIndex >= options.numTestingImages:
break
continue
if 'inference' not in options.dataset:
options.keyname = name
printStatisticsDetection(options, statistics)
all_statistics.append(statistics)
pass
continue
if 'inference' not in options.dataset:
if options.debug and len(detectors) > 1:
all_statistics = np.concatenate([
np.arange(len(all_statistics[0][0])).reshape((-1, 1)),
] + [np.array(statistics[3]) for statistics in all_statistics],
axis=-1)
print(all_statistics.astype(np.int32))
pass
if options.testingIndex == -1:
np.save('logs/all_statistics.npy', all_statistics)
pass
pass
return
# Root directory of the project
ROOT_DIR = os.path.abspath("../")
sys.path.append(ROOT_DIR) # To find local version of the library
# Logging confg
logging.basicConfig(level=logging.DEBUG,handlers=[
logging.FileHandler("{0}/{1}.log".format("/logs", "classifierservice-flask")),
logging.StreamHandler()])
############################################################
# Configurations
# Inherits from config.py
############################################################
from config import InferenceConfig
config = InferenceConfig()
# Create model object in inference mode.
module = __import__(config.MODEL_MODULE, fromlist=[config.MODEL_CLASS])
my_class = getattr(module,config.MODEL_CLASS)
model = my_class(config)
#Make a prediction before starting the server (First prediction takes longer)
data=config.MODEL_SAMPLE_INPUT
classification=model.predict(data)
logging.info('Model and weight have been loaded.')
def run_infer_content(data):
#logging.info('Load data: %s',data)
if isinstance(data,str):
###############################################################################
# Load image
###############################################################################
im_path = str(TRAIN_IMAGES_DIR / args.image)
print(f"Running on image '{args.image}'.")
original_im = cv2.imread(im_path)
original_im = cv2.cvtColor(original_im, cv2.COLOR_BGR2RGB)
im = original_im.copy()
###############################################################################
# Detection
###############################################################################
steel_config = InferenceConfig()
model = MaskRCNN(mode="inference",
config=steel_config,
model_dir=str(MODEL_DIR))
# Run the detection pipeline
# images: List of images, potentially of different sizes.
# Returns a list of dicts, one dict per image. The dict contains:
# rois : [N, (y1, x1, y2, x2)] detection bounding boxes
# class_ids : [N] int class IDs
# scores : [N] float probability scores for the class IDs
# masks : [H, W, N] instance binary masks
results = model.detect(images=[im], verbose=1)
r = results[0]
def __init__(self):
super().__init__()
model_config = InferenceConfig()
self.preprocess_obj = ForwardModel(model_config)
def __init__(self, config=None):
if config == None:
config = InferenceConfig()